PENERAPAN TEMPORAL FUSION TRANSFORMER (TFT) UNTUK PREDIKSI DATA MULTIVARIATE DERET WAKTU: STUDI KASUS HARGA SAHAM BBCA, BBRI, BMRI, BBNI, DAN BRIS

Agung Rashif Madani, NIM.: 20106050083 (2024) PENERAPAN TEMPORAL FUSION TRANSFORMER (TFT) UNTUK PREDIKSI DATA MULTIVARIATE DERET WAKTU: STUDI KASUS HARGA SAHAM BBCA, BBRI, BMRI, BBNI, DAN BRIS. Skripsi thesis, UIN SUNAN KALIJAGA YOGYAKARTA.

Preview	Text (PENERAPAN TEMPORAL FUSION TRANSFORMER (TFT) UNTUK PREDIKSI DATA MULTIVARIATE DERET WAKTU: STUDI KASUS HARGA SAHAM BBCA, BBRI, BMRI, BBNI, DAN BRIS) 20106050083_BAB-I_IV-atau-V_DAFTAR-PUSTAKA.pdf - Published Version Download (2MB) | Preview
	Text (PENERAPAN TEMPORAL FUSION TRANSFORMER (TFT) UNTUK PREDIKSI DATA MULTIVARIATE DERET WAKTU: STUDI KASUS HARGA SAHAM BBCA, BBRI, BMRI, BBNI, DAN BRIS) 20106050083_BAB-II_sampai_SEBELUM-BAB-TERAKHIR.pdf - Published Version Restricted to Registered users only Download (3MB) | Request a copy

Abstract

Technology and increased data volume drive Time Series Forecasting research, which is important in stock analysis for competitive advantage. However, challenges such as determining sufficient information and potential bias in model selection need to be addressed. Transformer-based Deep Learning models such as Temporal Fusion Transformer (TFT) have shown high effectiveness in handling multivariate time series data. Nonetheless, comprehensive studies are still needed in the implementation and evaluation of Transformer-based models such as TFT specifically for multivariate time series data prediction in the context of the financial industry. To overcome this shortcoming, this study proposes the use of TFT to determine the best model hyperparameter configuration and data preprocessing strategy in predicting the stock prices of the five largest banks in Indonesia (BBCA, BBRI, BMRI, BBNI, BRIS). This study considers two types of encoder-decoder, namely 5-1 and 25-5, two data normalization techniques, namely Standard Scaler and Robust Scaler, and three different batch sizes, namely 32, 64, and 128. A total of 12 combinations of TFT model hyperparameters were explored. The results showed that the optimal combination of data preprocessing, the use of a 25-5 encoder-decoder, normalization with Standard Scaler, and a batch size of 32. The evaluation results were MAE of 31.9601, RMSE of 40.1071, and SMAPE of 0.0074.

Share this knowledge with your friends :

Actions (login required)

View Item